Natural selection inference methods often target one mode of selection of a particular age and strength. However, detecting multiple modes simultaneously, or with atypical representations, would be advantageous for understanding a population’s evolutionary history. We have developed an anomaly detection algorithm using distributions of pairwise time to most recent common ancestor (TMRCA) to simultaneously detect multiple modes of natural selection in whole-genome sequences. Since natural selection distorts local genealogies in distinct ways, the method uses pairwise TMRCA distributions, which approximate genealogies at a non-recombining locus, to detect distortions without targeting a specific mode of selection. We evaluate the performance of our method, TSel, for both positive and balancing selection over different time-scales and selection strengths and compare TSel’s performance to that of other methods. We then apply TSel to the Complete Genomics diversity panel, a set of human whole-genome sequences, and recover loci previously inferred to be under positive or balancing selection.